Process for block quantizing an electrical signal and device for implementing said process

• US 4,051,470 A
• Filed: 03/31/1976
• Issued: 09/27/1977
• Est. Priority Date: 05/27/1975
• Status: Expired due to Term

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Esteban, Daniel J., Menez, Jean E.
• Primary Examiner(s): Miller, Charles D.
• Application Number:
  US05/672,203
• Time in Patent Office: 545 Days
• Field of Search: 325/38 R, 325/38 B, 325/42, 340/347 AD
• US Class Current: 341/118
• CPC Class Codes: H03M 3/022 with adaptable step size, e.g. adaptive delta modulation [ADM]
• CPC Class Codes: H03M 7/3053 Block-companding PCM systems

Abstract

This is a block quantizer which converts a given block of samples into an optimum sequence of numbers describing the sampled signal with minimum distortion due to the quantizing process.

The quantizer self-adjusts the basic parameters, i.e., quantizing step Q, d.c. level C and the sequence of numbers fn, describing the signal for each block of samples for a minimal Mean Squared Error.

Claims

1. A quantizing process for determining the components of a reference polynomial, a quantization parameter and a sequence of data bits for each section of a signal, each section being defined by a plurality of discrete information signals (Sn) with each signal being derived from the amplitude taken by said signal at an associated one of N points, said process being characterized by the steps of:

a. determining a sequence of data bits (Fn) wherein each bit represents a parameter of an associated one of the Sn'"'"'s;

b. determining the means quantization error of said sequence of data bits (Fn) for sqid signal of N points, said mean quantization error (E) satisfying the expression;

##STR1## in which p is a positive integer, Cn is a reference polynomial and Q is the quantization parameter for said Sn signals and,

c. the determination of new values of Cn and Q defining the minimum value of said mean quantizing error E, said determination including;

c1. the generation of a new sequence of data bits Fn using the values of Cn and Q derived from step (b) with the same plurality of discrete information signals (Sn) of said section;

c2. the generation of a new error E1 by using the newly generated values of Cn, Q and Fn, and

c3. the further repetition of steps (c1) and (c2) until a relatively fixed value of polynomial Cn is obtained.

2. A quantizing process according to claim 1, characterized in that the magnitude of said quantizing step may be modified in accordance with a predetermined rule during the execution of said iterative process for optimizing said parameter values Cn and Q to thereby minimize the value of said error E.

3. A plural bit quantizing process for a section of a data bearing signal, said section being defined by a sequence of numbers Fn, said process characterized in the N discrete information samples Sn comprising a section are split into 2^B -1 groups where B is the number of different quantizing parameters to be used and where a sample Sn may be included in either one or two of said groups, said process further characterized in that the quantization parameter in each of said groups is optimized in accordance with the process defined in claim 2, whereby the optimum values of the decision levels of the B bit coding are determined.

4. A bit quantizing device for determining a quantizing level which enables for information signal to be encoded with a minimal quantizing error, said device including the following elements:

a. means for storing a group of signal samples Sn (N ≧

n ≧

     1) derived from a section of an information signal to be quantized,

b. readout means connected to said storing means for making said samples available in sequence for a plurality of iterations,

c. means driven by said readout means for generating a first bit sequence Fn representing the successive arithmetic signs of said information samples Sn,

d. a computing means connected thereto for calculating from said samples Sn and said first bit sequence Fn, the values of the terms of a reference polynomial Cn and of a quantizing level Q which provides a relative minimum of the mean square error term due to the quantization which provides sequence Fn,

e. means for supplying said terms of said reference polynomial and a succeeding readout of said samples Sn from said storage means to said means driven by said readout means for determining a new sequence Fn representing the signs of the sequence of terms to generate a new quantizing level and,

f. further means for causing additional repetitions of the functions performed by means b, c, d, and e, until a relatively fixed reference polynomial Cn is obtained.

5. A device for quantizing in two bit form successive samples of a section of an information signal comprising a plurality of sections, each section being defined as a plurality of discrete information signals Sn, with each signal being derived from the amplitude taken by said information signal in a related one of N sampling intervals forming a signal section, characterized in that said device includes:

a. a means for determining a first sequence Fn of single bits representing the arithmetic sign of the successive Sn'"'"'s of a section,

b. a computing device connected to said means to receive both to said Fn and said Sn signals for generating an optimum 1-bit quantization by providing the values C of the terms of a reference polynomial Cn and evaluating a quantizing step Q,

c. means connected to said computing device for separating the Sn terms into three groups, i.e., those having values less than the value C-Q, those having values between C-Q and C+Q and those having values greater than C+Q,

d. a feedback means receiving the outputs of said separating means and connected to said computing device for then using said computing device to optimize the 1-bit quantization within each of said groups and for deriving the respective values of the corresponding terms C of reference polynomials C1, C2 and C3, one polynomial for each group,

e. additional control means set in accordance with said derived reference polynomials for again splitting said SN'"'"'s into three groups, the groups containing those Sn'"'"'s whose values are respectively less than C2, between C1 and C3 and higher than C2,

f. timing controls to control a plurality of repetitions of the operations performed by said feed back means and said additional control means until relatively fixed values are obtained for said reference polynomials C1, C2, and C3, and

g. means set thereby for then encoding each of the Sn'"'"'s in a two-bit form according to its value with respect to said values C1, C2 or C3, respectively.